Post on 02-Apr-2018
transcript
7/27/2019 Why Dont We See Poverty Convergence
1/46
Policy Research Working Paper 4974
W Dt W S Pvt Cv?
Martin Ravallion
WPS4974
PublicDisclosureAuthorized
PublicDisclosureAuthorized
ublicDisclosureAuthorized
7/27/2019 Why Dont We See Poverty Convergence
2/46
Abstract
Policy Research Working Paper 4974
W t t t vt mtt t dv ut. Yt t md b
wd td tzd t but t dvmt. T t t x wt mm t tzd t. Ctt wt md m wt t bw tt,t w dt t 100 dvut v dv t umtwt t vt d t v tm. A d vt t wubqut t t vt t v
T dut t Dt' O, Dvmt R Gu t t t dtmtt udtd w m ut m uu t ft t vt t t. P R WP td t Wb t tt://.wdb.. T ut m b ttd t mv@wdb..
wt t (d ut td t x t vt du ). Tu, m ut, t wt dvt ttut wt w m (dt v) tdu t t vt t. T z t mddmud b dv-ut, t Wt,tdd t b mtt ut vt t ubqut wt d vtdut. Hwv, ut qut d t t d vt tvt m umt.
7/27/2019 Why Dont We See Poverty Convergence
3/46
Why Dont We See Poverty Convergence?
Martin Ravallion1
Development Research Group, World Bank
1818 H Street NW, Washington DC, 20433, USA
7/27/2019 Why Dont We See Poverty Convergence
4/46
1. Introduction
Two widely-held stylized facts about development are (first) the advantage ofbackwardnessthe mean convergence property stemming from diminishing returns to
aggregate capital, such that we see higher growth rates in countries starting out with a low mean,
given otherwise similar initial conditionsand (second) the advantage of growthwhereby a
higher mean income tends to come with a lower incidence of absolute poverty; a stronger version
of the second stylized fact says that economic growth is uncorrelated with changes in
inequality. There seems to be ample evidence supporting both stylized facts.2
It appears to have gone unnoticed in the literature that, when taken together, these
stylized facts suggest that we should see poverty convergence: a catching up process whereby
countries starting out with a high incidence of poverty (reflecting a lower mean) will enjoy a
higher subsequent growth rate and (hence) higher pace of poverty reduction.
3
The possibility of dynamic efficiency costs of high current inequality has been raised in
the literature (reviewed later). But is it inequality or something else about the initial distribution
th t tt h t th i f th iddl l ? I lit i b i l t th
However, that
poses a puzzle since, as this paper will show, there is no sign of such poverty convergence; the
overall incidence of poverty is falling in the developing world, but no faster in its poorest
countries. Something is missing from these two stylized facts. Intuitively, it appears that either
the growth process in the mean, or the impact of growth on poverty, must depend directly on the
initial poverty rate, in a way that largely nullifies the advantage of backwardness.
7/27/2019 Why Dont We See Poverty Convergence
5/46
to help the middle-class may do little to relieve current poverty.) In fact many of the same
theories that have pointed to efficiency costs of high inequality also suggest that high poverty
handicaps growth.
To explore these issues empirically, a new data set was constructed for this paper from
household surveys for almost 100 developing countries, each with two or more surveys over
time. These data are used to estimate a model in which the rate of progress against poverty
depends on the rate of growth in the mean and various parameters of the initial distribution,
while the rate of growth depends in turn on initial distribution as well as the initial mean. The
model is subjected to a number of tests, including functional form, sample selection by type of
survey, and what measures are used for the key variables. A sub-sample with three or more
surveys is also used to test robustness to different specification choices, including treating initial
distribution as endogenous by treating lagged initial distribution as excludable.
The paper finds that mean-convergence is counteracted by two poverty effects. First,
there is an adverse direct effect of high initial poverty on growthworking against convergence
in mean incomes. Second, high initial poverty dulls the impact of growth on poverty; the poor
enjoy a lower share of the gains from growth in poorer countries.4 On balance there is little or no
systematic effect of starting out poor on the rate of poverty reduction. Other aspects of the initial
distribution play no more than a secondary role. High initial inequality only matters to growth
and poverty reduction in so far as it entails a high initial incidence of poverty relative to the
mean Countries starting out with a small middle classjudged by developing country rather
7/27/2019 Why Dont We See Poverty Convergence
6/46
2. Issues begging from past theories and evidence
There is a large literature, both theoretical and empirical, on growth and distributionalchange. Here I focus solely on the most relevant strands of that literature, and their implications
for the empirical work reported later.
2.1 Theories of distribution-dependent growth
A body of theoretical work has suggested that the initial distribution matters to an
economys aggregate efficiency and (hence) growth.
5
One class of models is based on the ideathat high inequality restricts efficiency-enhancing cooperation amongst people, such that key
public goods are underprovided or efficiency-enhancing policy reforms are blocked (Bardhan et
al., 2000). Other models argue that high inequality leads democratic governments to implement
distortionary redistributive policies; an example is the model of Alesina and Rodrik (1994).
Theories based on credit-market failures also suggest that inequality retards growth. The
market failure is typically attributed to information asymmetries, notably that lenders are
imperfectly informed about borrowers. The key analytic feature of this class of models is a
suitably nonlinear relationship between an individuals initial wealth and her future wealth (the
recursion diagram). The economic rationale for a nonlinear recursion diagram assumes that
credit market failures leave unexploited opportunities for investment in physical and human
capital. The information problem means that trades do not occur that would increase aggregate
output. With diminishing marginal products of capital, the mean future wealth will be a quasi-
7/27/2019 Why Dont We See Poverty Convergence
7/46
constraint implies that unambiguously higher current poverty incidencedefined by any poverty
line up to the minimum level of initial wealth needed to not be liquidity constrained in
investmentyields a lower growth at a given level of mean current wealth.
These are not the only arguments suggesting that poverty is a relevant parameter. Another
example can be found in the theories that have postulated impatience for consumption (high time
preference rates possibly associated with low life expectancy) and hence low savings and
investment rates by the poor (see, for example, Azariadis, 2006). Here too, while the theoretical
literature has focused on initial inequality, it can also be argued that a higher initial incidence of
poverty means a higher proportion of impatient consumers and hence lower growth.
Yet another example can be generated by considering how work productivity is affected
by past nutritional status and health status. Only when past nutritional intakes have been high
enough (above basal metabolic rate) will it be possible to do any work, but diminishing returns to
work will set in later; see the model in Dasgupta and Ray (1986). This type of argument can be
broadened to include other aspects of child development that have lasting impacts on learning
ability and earnings as an adult (Cunha and Heckman, 2007). Growing up in poverty can thus
have lasting impacts on aggregate efficiency.
There are also theoretical arguments involving market and institutional development,
though this is not a topic that has so far received as much attention in this literature. While past
theories have often taken credit-market failures to be exogenous, poverty may well be a deeper
causative factor in financial development (as well as an outcome of the lack of financial
7/27/2019 Why Dont We See Poverty Convergence
8/46
distribution-dependent growth suggest that the convergence effect is counterbalanced by an
adverse distributional effect of higher poverty. Which effect dominates is an empirical question.
A strand of the theoretical literature has also pointed to the possibilities for multiple
equilibria in nonlinear dynamic models, whereby one of the equilibria is a poverty trap (low-
level attractor). Essentially, the recursion diagram now has a low-level non-convexity, whereby
a minimum level of current wealth is essential before any positive level of future wealth can be
reached. In poor countries, the nutritional requirements for work can readily generate such
dynamics, as illustrated by the model of Dasgupta and Ray (1986). Such a model predicts that a
large exogenous income gain may be needed to attain a permanently higher income and that
seemingly similar aggregate shocks can have dissimilar outcomes; growth models with such
features are also discussed in Day (1992) and Azariades (1996, 2006) amongst others. Sachs
(2005) has invoked such models to argue that a large expansion of development aid would be
needed to assure a permanently higher average income in currently poor countries.
2.2 Implications of a growth model with borrowing constraints
Possibly the simplest illustration of how poverty can handicap growth is found in models
incorporating a borrowing constraint, whereby a person cannot borrow more than a fixed ratio of
their current wealth. This section draws out this implication from one such model.
Banerjee and Duflo (2003) provide a simple but insightful growth model with a
borrowing constraint. Someone who starts her productive life with sufficient wealth will invest
h t i d ti l t ti th (d li i ) i l d t f h it l ith
7/27/2019 Why Dont We See Poverty Convergence
9/46
strictly concave production function yielding output )(kh from a capital stockk. Given the rate
of interest r(taken to be fixed) the desired capital stock is
*
k , such that rkh = )(*
. Those with
initial wealth less than )1/(* +k will be credit constrained, in that after investing all they can
they will still find that rkh t > )( , while the rest will be free to implement*
k . A share
)1,0(1 of current wealth is consumed, leaving for the next period.
Under these assumptions, the recursion diagram takes the form:
]))1(([)(1 tttt rwwhww +==+ for )1/(* + kwt (1.1)
])()([ ** rkwkh t += for )1/(* +> kwt (1.2)
Plainly, )( tw is strictly concave up to )1/(* +k and linear above that. Mean future wealth is:
+ =0
1 )]([ dppwtt (2)
By well-known properties of concave functions, we have:
Proposition 1: (Banerjee and Duflo, 2003, p.277): An exogenous mean-preserving
spread in the wealth distribution in this economy will reduce future wealth and by
implication the growth rate.
However, the Banerjee-Duflo model has a further implication concerning poverty, asanother aspect of the initial distribution. Let )(zFH tt = denote the headcount index of poverty
(poverty rate) in this economy when the poverty line is z I assume that )1/(* + kz and let
7/27/2019 Why Dont We See Poverty Convergence
10/46
+
++=
+**
0
*
0
**
1 )()(])1))(()1(([tt H
t
t
H
t
tt
t
t dpH
pwrdp
H
pwrpwh
H
(4)
The sign of (4) cannot be determined under the assumptions so far.8
=t
However, on imposing a
constant initial mean , equation (4) simplifies to:
0)(
)1]())()1(([
*
0
**
1